On the 1st (local time), Nvidia Chief Executive Officer Jensen Huang declared in a keynote at Computex 2026 at the Taipei Music Center in Taiwan that the AI industry has entered a full-fledged commercialization phase. Nvidia's strategy is to go beyond supplying high-performance chips and build an ecosystem in which software and hardware around the world run on the company's computing infrastructure. In the same vein, Huang put "Agentic AI," which reasons and acts on its own, and "Physical AI," which applies it to the real world such as robots and Autonomous Driving cars, at the forefront. The idea is that as AI use expands, more computing resources will be needed, which will lead to increased demand for Nvidia's infrastructure.
About 5,000 members of the press and global information technology (IT) officials gathered at the venue to check a milestone for the global IT industry. Entry procedures began at 9 a.m., two hours before the event. The area in front of the Taipei Music Center, the venue, was packed with people waiting, and competition for seats among global media and industry officials was intense.
At 11 a.m., as Huang, in a black leather jacket, took the stage, the audience applauded. Opening his speech, Huang said, "My parents and family are here today," and introduced his father, Sidney Huang, his mother, and his eldest daughter, Madison Huang, seated in the audience.
In Korea's business community, Chey Tae-won, chairman of SK Group, drew attention by attending. Chey sat in the center seat of the front-row VIP section and watched the speech, and he took his seat alongside key Taiwanese partners, including MediaTek CEO Rick Tsai as well as executives from Foxconn and Asus, making a global AI alliance front visible.
Huang explained the mechanics of "token economics," in which computational efficiency consolidates into corporations' direct sales as Agentic AI enters the commercialization phase. As an indicator, he presented data from the software development platform GitHub and said that since the introduction of agent AI, the volume of code production (commits) by developers worldwide surged about threefold year over year as of early 2026. Addressing some concerns about job losses, he called them "complete nonsense," noting that the explosion in productivity instead creates a virtuous cycle that expands corporations' demand to hire engineers.
As a result, as corporations secure computational efficiency, the output generated by AI itself is converted into a direct revenue unit (token). That is why Huang repeatedly said during the speech, "The more you buy, the more you make," emphasizing the quantitative economics of investing in the company's infrastructure.
The next-generation infrastructure platform "Vera Rubin," which underpins this "AI factory," has already entered full production. Huang said, "The supply chain for Vera Rubin is twice as large as the previous generation, Grace Blackwell," and officially listed the memory supply chain, saying, "HBM4 memory will be supplied by SK hynix, Samsung Electronics, and Micron." It effectively reaffirmed on the global stage Nvidia's strategic partnerships with Korean companies in HBM4, the core of the next-generation accelerator market.
The "VERA CPU," unveiled alongside it, is Nvidia's key new product aimed at a server market centered on Intel and AMD. Huang said, "Legacy CPUs were designed in seconds for human users, but the Vera CPU is a processor redesigned from the ground up solely for AI agents that operate in nanoseconds." The strategy is to fundamentally block the compute bottlenecks that occur when agents access databases and use tools, pushing the utilization of the company's GPUs and token productivity to the extreme.
Nvidia then unveiled the open software model "Nemotron 3 Ultra" to help corporations build their own agents. At the same time, Nvidia's AI PC strategy took shape. Huang unveiled the on-device AI chips "RTX Spark" and "N1X," developed in partnership with Microsoft and MediaTek, formalizing its entry into the AI laptop market.
He said, "We will reinvent the PC with Microsoft," presenting a new PC environment in which AI agents run at all times in personal settings. He also predicted that the so-called "local AI (On-device AI)" market, which runs AI on the device without going through cloud servers, will expand in earnest.
The latter part of the speech focused on "Physical AI" solutions that merge with the physical world. Nvidia successively unveiled the foundation model for robots "Cosmos 3," the world's first inference-type open model for Autonomous Driving vehicles "AlphaMayo 2," and the reference humanoid platform "Isaac Groot." In the case of AlphaMayo 2, it will be applied to Nvidia's Hyperion ecosystem, which includes Mercedes-Benz and 80% of global carmakers, and the company plans to expand the architecture beyond virtual simulation to actual mobility and the broader robot factory infrastructure.
Partnerships with Korean corporations were also highlighted from multiple angles. Huang prominently named SK hynix and Samsung Electronics as HBM4 suppliers, and directly cited Naver Cloud, Hyundai Motor, and SKT as global partners. It shows that Nvidia's AI platform has taken root as core infrastructure across Korea's key value chain, from high-performance memory supply to the cloud, and on to traditional manufacturing and mobility.
Nvidia presented a technology roadmap that connects data center infrastructure spanning accelerators (GPUs), dedicated CPUs, and key memory with AI PCs, On-device AI, Robotics, and Autonomous Driving. If last year's Computex was a stage to persuade on the need for data center expansion and hardware infrastructure build-out, this year the focus was on completing a practical commercialization software roadmap that will run on that infrastructure.